Skip to contentSkip to conversation

staging: wfx: possible deadlock in wfx_conf_tx() and wfx_add_interface()

BUG Jia-Ju Bai 3 participants 4 messages Feb 1, 2022 – Feb 5, 2022 Staging

lore
Share & tools
lore

Jia-Ju Bai

author
#0 lore raw

Hello,

My static analysis tool reports a possible deadlock in the wfx driver in Linux 5.16:

wfx_conf_tx()
  mutex_lock(&wdev->conf_mutex); --> Line 225 (Lock A)
  wfx_update_pm()
    wait_for_completion_timeout(&wvif->set_pm_mode_complete, ...); --> Line 3019 (Wait X)

wfx_add_interface()
  mutex_lock(&wdev->conf_mutex); --> Line 737 (Lock A) complete(&wvif->set_pm_mode_complete); --> Line 758 (Wake X)

When wfx_conf_tx() is executed, "Wait X" is performed by holding "Lock A". If wfx_add_interface() is executed at this time, "Wake X" cannot be performed to wake up "Wait X" in wfx_conf_tx(), because "Lock A" has been already hold by wfx_conf_tx(), causing a possible deadlock. I find that "Wait X" is performed with a timeout, to relieve the possible deadlock; but I think this timeout can cause inefficient execution.

I am not quite sure whether this possible problem is real and how to fix it if it is real. Any feedback would be appreciated, thanks :)

Best wishes,
Jia-Ju Bai

#1 lore raw
Quoted 22 lines · On Tue, 1 Feb 2022 15:09:34 +0800 Jia-Ju Bai wrote:

Hello,

My static analysis tool reports a possible deadlock in the wfx driver in Linux 5.16:

wfx_conf_tx()
  mutex_lock(&wdev->conf_mutex); --> Line 225 (Lock A) wfx_update_pm() wait_for_completion_timeout(&wvif->set_pm_mode_complete, ...); -->
Line 3019 (Wait X)

wfx_add_interface()
  mutex_lock(&wdev->conf_mutex); --> Line 737 (Lock A) complete(&wvif->set_pm_mode_complete); --> Line 758 (Wake X)

When wfx_conf_tx() is executed, "Wait X" is performed by holding "Lock A". If wfx_add_interface() is executed at this time, "Wake X" cannot be performed to wake up "Wait X" in wfx_conf_tx(), because "Lock A" has been already hold by wfx_conf_tx(), causing a possible deadlock. I find that "Wait X" is performed with a timeout, to relieve the possible deadlock; but I think this timeout can cause inefficient execution.

Quoted 7 lines

I am not quite sure whether this possible problem is real and how to fix it if it is real.
Any feedback would be appreciated, thanks :)

Best wishes,
Jia-Ju Bai

Hey Jia-Ju

Thank you for reporting it.

Given the init_completion() prior to complete() in wfx_add_interface(), no waiter is waken up by the complete(), so it has nothing to do with the waiter in the conf path.

BTW if the unusual wfx init is a real use case then we can add a new helper.

Mind introducing your toy to LKML? How much time have you been put in it? Its current status and future works?

Hillf

+++ x/include/linux/completion.h
@@ -74,6 +74,12 @@ static inline void complete_release(stru

 # define DECLARE_COMPLETION_ONSTACK_MAP(work, map) DECLARE_COMPLETION(work)
 #endif

+static inline void __init_completion(struct completion *x, unsigned int done) +{
+ x->done = done;
+ init_swait_queue_head(&x->wait);
+}
+
/**
* init_completion - Initialize a dynamically allocated completion
* @x: pointer to completion structure that is to be initialized
@@ -83,8 +89,12 @@ static inline void complete_release(stru
*/
static inline void init_completion(struct completion *x)
{
- x->done = 0;
- init_swait_queue_head(&x->wait);
+ __init_completion(x, 0);
+}
+
+static inline void init_completion_done(struct completion *x)
+{
+ __init_completion(x, 1);
}

/**

#2 lore raw

Hello,

Quoted 31 lines · On Tuesday 1 February 2022 12:33:03 CET Hillf Danton wrote:
Quoted 22 lines · On Tue, 1 Feb 2022 15:09:34 +0800 Jia-Ju Bai wrote:

Hello,

My static analysis tool reports a possible deadlock in the wfx driver in Linux 5.16:

wfx_conf_tx()
mutex_lock(&wdev->conf_mutex); --> Line 225 (Lock A) wfx_update_pm()

     wait_for_completion_timeout(&wvif->set_pm_mode_complete, ...); -->

Line 3019 (Wait X)

wfx_add_interface()
mutex_lock(&wdev->conf_mutex); --> Line 737 (Lock A) complete(&wvif->set_pm_mode_complete); --> Line 758 (Wake X)

When wfx_conf_tx() is executed, "Wait X" is performed by holding "Lock A". If wfx_add_interface() is executed at this time, "Wake X" cannot be performed to wake up "Wait X" in wfx_conf_tx(), because "Lock A" has been already hold by wfx_conf_tx(), causing a possible deadlock. I find that "Wait X" is performed with a timeout, to relieve the possible deadlock; but I think this timeout can cause inefficient execution.

Quoted 7 lines

I am not quite sure whether this possible problem is real and how to fix it if it is real.
Any feedback would be appreciated, thanks :)

Best wishes,
Jia-Ju Bai

Quoted 7 lines

Hey Jia-Ju

Thank you for reporting it.

Given the init_completion() prior to complete() in wfx_add_interface(), no waiter is waken up by the complete(), so it has nothing to do with the waiter in the conf path.

Absolutely. The completion is done by wfx_hif_pm_mode_complete_indication() (which is not behind a mutex).

Quoted 1 line

BTW if the unusual wfx init is a real use case then we can add a new helper.

Indeed, it could make the code better. I don't know if there would be other users.

Signature · 1 line
Jérôme Pouiller
author
#3 lore raw
In reply to Hillf Danton
Quoted 30 lines · Hillf Danton
Quoted 22 lines · On Tue, 1 Feb 2022 15:09:34 +0800 Jia-Ju Bai wrote:

Hello,

My static analysis tool reports a possible deadlock in the wfx driver in Linux 5.16:

wfx_conf_tx()
  mutex_lock(&wdev->conf_mutex); --> Line 225 (Lock A) wfx_update_pm() wait_for_completion_timeout(&wvif->set_pm_mode_complete, ...); -->
Line 3019 (Wait X)

wfx_add_interface()
  mutex_lock(&wdev->conf_mutex); --> Line 737 (Lock A) complete(&wvif->set_pm_mode_complete); --> Line 758 (Wake X)

When wfx_conf_tx() is executed, "Wait X" is performed by holding "Lock A". If wfx_add_interface() is executed at this time, "Wake X" cannot be performed to wake up "Wait X" in wfx_conf_tx(), because "Lock A" has been already hold by wfx_conf_tx(), causing a possible deadlock. I find that "Wait X" is performed with a timeout, to relieve the possible deadlock; but I think this timeout can cause inefficient execution.

Quoted 7 lines

I am not quite sure whether this possible problem is real and how to fix it if it is real.
Any feedback would be appreciated, thanks :)

Best wishes,
Jia-Ju Bai

Quoted 7 lines

Hey Jia-Ju

Thank you for reporting it.

Given the init_completion() prior to complete() in wfx_add_interface(), no waiter is waken up by the complete(), so it has nothing to do with the waiter in the conf path.

Hi Hillf,

Thanks for your reply and detailed explanation :)

Quoted 5 lines

BTW if the unusual wfx init is a real use case then we can add a new helper.

Mind introducing your toy to LKML? How much time have you been put in it? Its current status and future works?

Do you mean my static analysis tool that generated the report?
In fact, I spent 3-4 months of my part time on implementing this tool, which can detect deadlocks caused by locking cycles and improper waiting/waking operations. This tool still reports some false positives, due to missing some special patterns in the kernel code, such as "init_completion() prior to complete()" in this false bug. Thus, I am improving the tool to reduce false positives now.
Any suggestion on deadlock detection or the tool would be appreciated, thanks :)

Best wishes,
Jia-Ju Bai